(00:15:16) alexr: ... but it's looking like -- and this is kind of interesting -- using an HMM is actually slightly worse than always just taking the most-frequent-translation.
(00:16:04) [jonathan]: HMM?
(00:16:08) alexr: hidden markov model.
(00:16:36) [jonathan]: ah
(00:17:14) [jonathan]: hrm
(00:17:17) alexr: No, hmm.